Original title: microsoft/VibeVoice
Microsoft's open-source family of speech models: VibeVoice-ASR can process long audio up to 60 minutes in a single pass, outputting structured transcriptions with speaker identification and timestamps, supporting 50 languages and custom contexts, and has been integrated into Hugging Face Transformers; ASR-Streaming, released in 9, enables real-time transcription as audio is captured; VibeVoice-Realtime-0.5B is a real-time speech synthesis model that supports streaming text input.
Stars 55952 (+223 / 24h)
Reasons
- Gained 223 stars in the last 24 hours, totaling 55952 stars, ranked #20 on Trendshift for the day (reflecting attention only).
Risks
- The code for long-form multi-speaker speech synthesis (VibeVoice-TTS) has been removed from the repository due to misuse.
- The README warns that synthesized speech may be used for impersonation and fraud, so usage must be lawful and clearly labeled as AI-generated.
- The most recent update in the README is from 9, so this listing does not indicate a new release.